AI chip designer AMD will procure HBM4 memory from Samsung and has signed an agreement to this effect, according to remarks made by CEO Lisa Su. The firm held its Advancing AI event earlier this week, and Su revealed the decision to bring Samsung into the fold to reporters at the conference. For memory manufacturers, securing contracts from firms such as AMD comes after they have qualified their products to meet the designers' specifications. At the event, Su announced AMD's Helios AI rack solutions to further beef up its portfolio and compete with NVIDIA.
AMD Is Close To Procure Large Amounts Of HBM4 Memory Chips From Samsung, Says CEO
AMD's Helios system is a key announcement as it is the firm's first foray into delivering rack-scale systems for AI data center computing. These systems combine the firm's networking systems, its AI GPUs, software and CPUs. The system was first discussed last the firm's Advancing AI event in 2025, and it is designed to compete with NVIDIA's Oberon and Kyber products.
It fuses together AMD's Instinct MI455X GPUs, its 6th Gen AMD EPYC CPUs and the AMD Pensando AI NICs. Additionally, the firm's Pensando DPU, Infinity Fabric and the ROCm software stack also play a role.
During her address at the AI event, Su confirmed that Helios was in full-scale production and was set to begin shipping by the end of this year's third quarter. Following the event, while she was talking with journalists, the AMD CEO also confirmed that her firm was on its way to sign an agreement to procure memory chips from Samsung, says a report from the Korean media. The CEO had visited Samsung's Pyeongtaek campus to sign a memorandum of understanding (MOU) in March.
The MOU is for priority supply for memory chips, and it came at a time when the global memory industry is undergoing a historic reshaping following high demand from AI chips. In response to media queries at the event, Su remarked that AMD was 'almost there' when it came to signing an agreement to procure HBM4 memory from Samsung.
The Korean firm's executives were less forthcoming about an agreement as they simply shared that they had an important meeting with AMD coming up.
AMD's Instinct MI455X GPU relies on 50% more HBM4 memory to compete with NVIDIA's Rubin and deliver 40 PFLOPs of AI computing capacity. The chip packs 320 billion transistors, and it is built with the 2-nanometer and 3-nanometer chip manufacturing technologies.
About the author: Ramish is a seasoned technology writer and editor with more than a decade of experience. He specializes in semiconductor fabrication and market analysis. With a background in finance and supply chain management - via his bachelors in Finance and a micromasters in supply chain management from MIT - Ramish combines financial rigor with deep industry insight to deliver accurate and authoritative coverage.
Follow Wccftech on Google to get more of our news coverage in your feeds.
